sbin/hammer/ondisk.c

   1 /*
   2  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  */
  34
  35 #include <sys/diskslice.h>
  36 #include <sys/diskmbr.h>
  37
  38 #include "hammer_util.h"
  39
  40 static void get_buffer_readahead(struct buffer_info *base);
  41 static void *get_ondisk(hammer_off_t buf_offset, struct buffer_info **bufferp,
  42                         int isnew);
  43 static __inline int readhammervol(struct volume_info *vol);
  44 static __inline int readhammerbuf(struct buffer_info *buf);
  45 static __inline int writehammervol(struct volume_info *vol);
  46 static __inline int writehammerbuf(struct buffer_info *buf);
  47
  48 int DebugOpt;
  49
  50 uuid_t Hammer_FSType;
  51 uuid_t Hammer_FSId;
  52 int     UseReadBehind = -4;
  53 int     UseReadAhead = 4;
  54 int     AssertOnFailure = 1;
  55 struct volume_list VolList = TAILQ_HEAD_INITIALIZER(VolList);
  56 static int valid_hammer_volumes;
  57
  58 static __inline
  59 int
  60 buffer_hash(hammer_off_t buf_offset)
  61 {
  62         int hi;
  63
  64         hi = (int)(buf_offset / HAMMER_BUFSIZE) & HAMMER_BUFLISTMASK;
  65         return(hi);
  66 }
  67
  68 static struct buffer_info*
  69 find_buffer(struct volume_info *volume, hammer_off_t buf_offset)
  70 {
  71         int hi;
  72         struct buffer_info *buf;
  73
  74         hi = buffer_hash(buf_offset);
  75         TAILQ_FOREACH(buf, &volume->buffer_lists[hi], entry)
  76                 if (buf->buf_offset == buf_offset)
  77                         return(buf);
  78         return(NULL);
  79 }
  80
  81 static
  82 struct volume_info *
  83 __alloc_volume(const char *volname, int oflags)
  84 {
  85         struct volume_info *vol;
  86         int i;
  87
  88         vol = malloc(sizeof(*vol));
  89         if (vol == NULL)
  90                 err(1, "alloc_volume");
  91         bzero(vol, sizeof(*vol));
  92
  93         vol->vol_no = -1;
  94         vol->rdonly = (oflags == O_RDONLY);
  95         vol->name = strdup(volname);
  96         vol->fd = open(vol->name, oflags);
  97         if (vol->fd < 0)
  98                 err(1, "alloc_volume: Failed to open %s", vol->name);
  99
 100         vol->size = 0;
 101         vol->device_offset = 0;
 102         vol->type = NULL;
 103
 104         vol->ondisk = malloc(HAMMER_BUFSIZE);
 105         if (vol->ondisk == NULL)
 106                 err(1, "alloc_volume");
 107         bzero(vol->ondisk, HAMMER_BUFSIZE);
 108
 109         for (i = 0; i < HAMMER_BUFLISTS; ++i)
 110                 TAILQ_INIT(&vol->buffer_lists[i]);
 111
 112         return(vol);
 113 }
 114
 115 static void
 116 __add_volume(struct volume_info *vol)
 117 {
 118         struct volume_info *scan;
 119         struct stat st1, st2;
 120
 121         if (fstat(vol->fd, &st1) != 0)
 122                 errx(1, "add_volume: %s: Failed to stat", vol->name);
 123
 124         TAILQ_FOREACH(scan, &VolList, entry) {
 125                 if (scan->vol_no == vol->vol_no) {
 126                         errx(1, "add_volume: %s: Duplicate volume number %d "
 127                                 "against %s",
 128                                 vol->name, vol->vol_no, scan->name);
 129                 }
 130                 if (fstat(scan->fd, &st2) != 0) {
 131                         errx(1, "add_volume: %s: Failed to stat %s",
 132                                 vol->name, scan->name);
 133                 }
 134                 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) {
 135                         errx(1, "add_volume: %s: Specified more than once",
 136                                 vol->name);
 137                 }
 138         }
 139
 140         TAILQ_INSERT_TAIL(&VolList, vol, entry);
 141 }
 142
 143 /*
 144  * Initialize a volume structure and ondisk vol_no field.
 145  */
 146 struct volume_info *
 147 init_volume(int32_t vol_no, const char *filename, int oflags)
 148 {
 149         struct volume_info *vol;
 150
 151         vol = __alloc_volume(filename, oflags);
 152         vol->vol_no = vol->ondisk->vol_no = vol_no;
 153
 154         __add_volume(vol);
 155
 156         return(vol);
 157 }
 158
 159 /*
 160  * Initialize a volume structure and read ondisk volume header.
 161  */
 162 struct volume_info*
 163 load_volume(const char *filename, int oflags)
 164 {
 165         struct volume_info *vol;
 166         struct hammer_volume_ondisk *ondisk;
 167         int n;
 168
 169         vol = __alloc_volume(filename, oflags);
 170
 171         n = readhammervol(vol);
 172         if (n == -1) {
 173                 err(1, "load_volume: %s: Read failed at offset 0", vol->name);
 174         }
 175         ondisk = vol->ondisk;
 176         vol->vol_no = ondisk->vol_no;
 177
 178         if (ondisk->vol_rootvol != HAMMER_ROOT_VOLNO) {
 179                 errx(1, "load_volume: Invalid root volume# %d",
 180                         ondisk->vol_rootvol);
 181         }
 182
 183         if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType))) {
 184                 errx(1, "load_volume: %s: Header does not indicate "
 185                         "that this is a hammer volume", vol->name);
 186         }
 187
 188         if (valid_hammer_volumes++ == 0) {
 189                 Hammer_FSId = ondisk->vol_fsid;
 190         } else if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId))) {
 191                 errx(1, "load_volume: %s: FSId does match other volumes!",
 192                         vol->name);
 193         }
 194
 195         __add_volume(vol);
 196
 197         return(vol);
 198 }
 199
 200 /*
 201  * Check basic volume characteristics.
 202  */
 203 void
 204 check_volume(struct volume_info *vol)
 205 {
 206         struct partinfo pinfo;
 207         struct stat st;
 208
 209         /*
 210          * Get basic information about the volume
 211          */
 212         if (ioctl(vol->fd, DIOCGPART, &pinfo) < 0) {
 213                 /*
 214                  * Allow the formatting of regular files as HAMMER volumes
 215                  */
 216                 if (fstat(vol->fd, &st) < 0)
 217                         err(1, "Unable to stat %s", vol->name);
 218                 vol->size = st.st_size;
 219                 vol->type = "REGFILE";
 220         } else {
 221                 /*
 222                  * When formatting a block device as a HAMMER volume the
 223                  * sector size must be compatible.  HAMMER uses 16384 byte
 224                  * filesystem buffers.
 225                  */
 226                 if (pinfo.reserved_blocks) {
 227                         errx(1, "HAMMER cannot be placed in a partition "
 228                                 "which overlaps the disklabel or MBR");
 229                 }
 230                 if (pinfo.media_blksize > HAMMER_BUFSIZE ||
 231                     HAMMER_BUFSIZE % pinfo.media_blksize) {
 232                         errx(1, "A media sector size of %d is not supported",
 233                              pinfo.media_blksize);
 234                 }
 235
 236                 vol->size = pinfo.media_size;
 237                 vol->device_offset = pinfo.media_offset;
 238                 vol->type = "DEVICE";
 239         }
 240 }
 241
 242 struct volume_info *
 243 get_volume(int32_t vol_no)
 244 {
 245         struct volume_info *vol;
 246
 247         TAILQ_FOREACH(vol, &VolList, entry) {
 248                 if (vol->vol_no == vol_no)
 249                         break;
 250         }
 251         if (vol == NULL)
 252                 errx(1, "get_volume: Volume %d does not exist!", vol_no);
 253
 254         /* not added to or removed from hammer cache */
 255         return(vol);
 256 }
 257
 258 struct volume_info *
 259 get_root_volume(void)
 260 {
 261         return(get_volume(HAMMER_ROOT_VOLNO));
 262 }
 263
 264 void
 265 rel_volume(struct volume_info *volume __unused)
 266 {
 267         /* nothing to do */
 268 }
 269
 270 /*
 271  * Acquire the specified buffer.  isnew is -1 only when called
 272  * via get_buffer_readahead() to prevent another readahead.
 273  */
 274 struct buffer_info *
 275 get_buffer(hammer_off_t buf_offset, int isnew)
 276 {
 277         struct buffer_info *buf;
 278         struct volume_info *volume;
 279         int vol_no;
 280         int zone;
 281         int hi, n;
 282         int dora = 0;
 283
 284         zone = HAMMER_ZONE_DECODE(buf_offset);
 285         if (zone > HAMMER_ZONE_RAW_BUFFER_INDEX) {
 286                 buf_offset = blockmap_lookup(buf_offset, NULL, NULL, NULL);
 287         }
 288         if (buf_offset == HAMMER_OFF_BAD)
 289                 return(NULL);
 290         assert(hammer_is_zone_raw_buffer(buf_offset));
 291
 292         vol_no = HAMMER_VOL_DECODE(buf_offset);
 293         volume = get_volume(vol_no);
 294
 295         buf_offset &= ~HAMMER_BUFMASK64;
 296         buf = find_buffer(volume, buf_offset);
 297
 298         if (buf == NULL) {
 299                 buf = malloc(sizeof(*buf));
 300                 bzero(buf, sizeof(*buf));
 301                 buf->buf_offset = buf_offset;
 302                 buf->raw_offset = hammer_xlate_to_phys(volume->ondisk,
 303                                                         buf_offset);
 304                 buf->volume = volume;
 305                 buf->ondisk = malloc(HAMMER_BUFSIZE);
 306                 if (isnew <= 0) {
 307                         n = readhammerbuf(buf);
 308                         if (n == -1) {
 309                                 if (AssertOnFailure)
 310                                         err(1, "get_buffer: %s:%016jx "
 311                                             "Read failed at offset %016jx",
 312                                             volume->name,
 313                                             (intmax_t)buf->buf_offset,
 314                                             (intmax_t)buf->raw_offset);
 315                                 bzero(buf->ondisk, HAMMER_BUFSIZE);
 316                         }
 317                 }
 318
 319                 hi = buffer_hash(buf_offset);
 320                 TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buf, entry);
 321                 buf->cache.buffer = buf;
 322                 hammer_cache_add(&buf->cache);
 323                 dora = (isnew == 0);
 324         } else {
 325                 assert(buf->ondisk != NULL);
 326                 assert(isnew != -1);
 327                 hammer_cache_used(&buf->cache);
 328         }
 329
 330         ++buf->cache.refs;
 331         hammer_cache_flush();
 332
 333         if (isnew > 0) {
 334                 assert(buf->cache.modified == 0);
 335                 bzero(buf->ondisk, HAMMER_BUFSIZE);
 336                 buf->cache.modified = 1;
 337         }
 338         if (dora)
 339                 get_buffer_readahead(buf);
 340         return(buf);
 341 }
 342
 343 static void
 344 get_buffer_readahead(struct buffer_info *base)
 345 {
 346         struct buffer_info *buf;
 347         struct volume_info *vol;
 348         hammer_off_t buf_offset;
 349         int64_t raw_offset;
 350         int ri = UseReadBehind;
 351         int re = UseReadAhead;
 352
 353         raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE;
 354         vol = base->volume;
 355
 356         while (ri < re) {
 357                 if (raw_offset >= vol->ondisk->vol_buf_end)
 358                         break;
 359                 if (raw_offset < vol->ondisk->vol_buf_beg || ri == 0) {
 360                         ++ri;
 361                         raw_offset += HAMMER_BUFSIZE;
 362                         continue;
 363                 }
 364                 buf_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no,
 365                         raw_offset - vol->ondisk->vol_buf_beg);
 366                 buf = find_buffer(vol, buf_offset);
 367                 if (buf == NULL) {
 368                         buf = get_buffer(buf_offset, -1);
 369                         rel_buffer(buf);
 370                 }
 371                 ++ri;
 372                 raw_offset += HAMMER_BUFSIZE;
 373         }
 374 }
 375
 376 void
 377 rel_buffer(struct buffer_info *buffer)
 378 {
 379         struct volume_info *volume;
 380         int hi;
 381
 382         if (buffer == NULL)
 383                 return;
 384         assert(buffer->cache.refs > 0);
 385         if (--buffer->cache.refs == 0) {
 386                 if (buffer->cache.delete) {
 387                         hi = buffer_hash(buffer->buf_offset);
 388                         volume = buffer->volume;
 389                         if (buffer->cache.modified)
 390                                 flush_buffer(buffer);
 391                         TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
 392                         hammer_cache_del(&buffer->cache);
 393                         free(buffer->ondisk);
 394                         free(buffer);
 395                         rel_volume(volume);
 396                 }
 397         }
 398 }
 399
 400 /*
 401  * Retrieve a pointer to a buffer data given a buffer offset.  The underlying
 402  * bufferp is freed if isnew or the offset is out of range of the cached data.
 403  * If bufferp is freed a referenced buffer is loaded into it.
 404  */
 405 void *
 406 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp,
 407                 int isnew)
 408 {
 409         if (*bufferp != NULL) {
 410                 if (isnew > 0 ||
 411                     (((*bufferp)->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) {
 412                         rel_buffer(*bufferp);
 413                         *bufferp = NULL;
 414                 }
 415         }
 416         return(get_ondisk(buf_offset, bufferp, isnew));
 417 }
 418
 419 /*
 420  * Retrieve a pointer to a B-Tree node given a zone offset.  The underlying
 421  * bufferp is freed if non-NULL and a referenced buffer is loaded into it.
 422  */
 423 hammer_node_ondisk_t
 424 get_node(hammer_off_t node_offset, struct buffer_info **bufferp)
 425 {
 426         if (*bufferp != NULL) {
 427                 rel_buffer(*bufferp);
 428                 *bufferp = NULL;
 429         }
 430         return(get_ondisk(node_offset, bufferp, 0));
 431 }
 432
 433 /*
 434  * Return a pointer to a buffer data given a buffer offset.
 435  * If *bufferp is NULL acquire the buffer otherwise use that buffer.
 436  */
 437 static void *
 438 get_ondisk(hammer_off_t buf_offset, struct buffer_info **bufferp, int isnew)
 439 {
 440         struct buffer_info *buffer;
 441
 442         buffer = *bufferp;
 443         if (buffer == NULL) {
 444                 buffer = *bufferp = get_buffer(buf_offset, isnew);
 445                 if (buffer == NULL)
 446                         return(NULL);
 447         }
 448
 449         return((char *)buffer->ondisk +
 450                 ((int32_t)buf_offset & HAMMER_BUFMASK));
 451 }
 452
 453 /*
 454  * Allocate HAMMER elements - B-Tree nodes
 455  */
 456 void *
 457 alloc_btree_element(hammer_off_t *offp, struct buffer_info **data_bufferp)
 458 {
 459         hammer_node_ondisk_t node;
 460
 461         node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
 462                               offp, data_bufferp);
 463         bzero(node, sizeof(*node));
 464         return (node);
 465 }
 466
 467 /*
 468  * Allocate HAMMER elements - meta data (inode, direntry, PFS, etc)
 469  */
 470 void *
 471 alloc_meta_element(hammer_off_t *offp, int32_t data_len,
 472                    struct buffer_info **data_bufferp)
 473 {
 474         void *data;
 475
 476         data = alloc_blockmap(HAMMER_ZONE_META_INDEX, data_len,
 477                               offp, data_bufferp);
 478         bzero(data, data_len);
 479         return (data);
 480 }
 481
 482 /*
 483  * Allocate HAMMER elements - data storage
 484  *
 485  * The only data_len supported by HAMMER userspace for large data zone
 486  * (zone 10) is HAMMER_BUFSIZE which is 16KB.  >16KB data does not fit
 487  * in a buffer allocated by get_buffer().  Also alloc_blockmap() does
 488  * not consider >16KB buffer size.
 489  */
 490 void *
 491 alloc_data_element(hammer_off_t *offp, int32_t data_len,
 492                    struct buffer_info **data_bufferp)
 493 {
 494         void *data;
 495         int zone;
 496
 497         if (data_len == 0)
 498                 return(NULL);
 499
 500         zone = hammer_data_zone_index(data_len);
 501         assert(data_len <= HAMMER_BUFSIZE); /* just one buffer */
 502         assert(zone == HAMMER_ZONE_LARGE_DATA_INDEX ||
 503                zone == HAMMER_ZONE_SMALL_DATA_INDEX);
 504
 505         data = alloc_blockmap(zone, data_len, offp, data_bufferp);
 506         bzero(data, data_len);
 507         return(data);
 508 }
 509
 510 /*
 511  * Format a new blockmap.  This is mostly a degenerate case because
 512  * all allocations are now actually done from the freemap.
 513  */
 514 void
 515 format_blockmap(struct volume_info *root_vol, int zone, hammer_off_t offset)
 516 {
 517         hammer_blockmap_t blockmap;
 518         hammer_off_t zone_base;
 519
 520         /* Only root volume needs formatting */
 521         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
 522
 523         assert(hammer_is_zone2_mapped_index(zone));
 524
 525         blockmap = &root_vol->ondisk->vol0_blockmap[zone];
 526         zone_base = HAMMER_ZONE_ENCODE(zone, offset);
 527
 528         bzero(blockmap, sizeof(*blockmap));
 529         blockmap->phys_offset = 0;
 530         blockmap->first_offset = zone_base;
 531         blockmap->next_offset = zone_base;
 532         blockmap->alloc_offset = HAMMER_ENCODE(zone, 255, -1);
 533         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
 534 }
 535
 536 /*
 537  * Format a new freemap.  Set all layer1 entries to UNAVAIL.  The initialize
 538  * code will load each volume's freemap.
 539  */
 540 void
 541 format_freemap(struct volume_info *root_vol)
 542 {
 543         struct buffer_info *buffer = NULL;
 544         hammer_off_t layer1_offset;
 545         hammer_blockmap_t blockmap;
 546         struct hammer_blockmap_layer1 *layer1;
 547         int i, isnew;
 548
 549         /* Only root volume needs formatting */
 550         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
 551
 552         layer1_offset = alloc_bigblock(root_vol, HAMMER_ZONE_FREEMAP_INDEX);
 553         for (i = 0; i < HAMMER_BIGBLOCK_SIZE; i += sizeof(*layer1)) {
 554                 isnew = ((i % HAMMER_BUFSIZE) == 0);
 555                 layer1 = get_buffer_data(layer1_offset + i, &buffer, isnew);
 556                 bzero(layer1, sizeof(*layer1));
 557                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 558                 layer1->blocks_free = 0;
 559                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 560         }
 561         assert(i == HAMMER_BIGBLOCK_SIZE);
 562         rel_buffer(buffer);
 563
 564         blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 565         bzero(blockmap, sizeof(*blockmap));
 566         blockmap->phys_offset = layer1_offset;
 567         blockmap->first_offset = 0;
 568         blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
 569         blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
 570         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
 571 }
 572
 573 /*
 574  * Load the volume's remaining free space into the freemap.
 575  *
 576  * Returns the number of big-blocks available.
 577  */
 578 int64_t
 579 initialize_freemap(struct volume_info *vol)
 580 {
 581         struct volume_info *root_vol;
 582         struct buffer_info *buffer1 = NULL;
 583         struct buffer_info *buffer2 = NULL;
 584         struct hammer_blockmap_layer1 *layer1;
 585         struct hammer_blockmap_layer2 *layer2;
 586         hammer_off_t layer1_offset;
 587         hammer_off_t layer2_offset;
 588         hammer_off_t phys_offset;
 589         hammer_off_t block_offset;
 590         hammer_off_t aligned_vol_free_end;
 591         hammer_blockmap_t freemap;
 592         int64_t count = 0;
 593         int64_t layer1_count = 0;
 594
 595         root_vol = get_root_volume();
 596         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
 597                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
 598
 599         printf("initialize freemap volume %d\n", vol->vol_no);
 600
 601         /*
 602          * Initialize the freemap.  First preallocate the big-blocks required
 603          * to implement layer2.   This preallocation is a bootstrap allocation
 604          * using blocks from the target volume.
 605          */
 606         freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 607
 608         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 609              phys_offset < aligned_vol_free_end;
 610              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 611                 layer1_offset = freemap->phys_offset +
 612                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 613                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 614                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
 615                         layer1->phys_offset = alloc_bigblock(vol,
 616                                                 HAMMER_ZONE_FREEMAP_INDEX);
 617                         layer1->blocks_free = 0;
 618                         buffer1->cache.modified = 1;
 619                         layer1->layer1_crc = crc32(layer1,
 620                                                    HAMMER_LAYER1_CRCSIZE);
 621                 }
 622         }
 623
 624         /*
 625          * Now fill everything in.
 626          */
 627         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 628              phys_offset < aligned_vol_free_end;
 629              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 630                 layer1_count = 0;
 631                 layer1_offset = freemap->phys_offset +
 632                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 633                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 634                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
 635
 636                 for (block_offset = 0;
 637                      block_offset < HAMMER_BLOCKMAP_LAYER2;
 638                      block_offset += HAMMER_BIGBLOCK_SIZE) {
 639                         layer2_offset = layer1->phys_offset +
 640                                         HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
 641                         layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
 642                         bzero(layer2, sizeof(*layer2));
 643
 644                         if (phys_offset + block_offset < vol->vol_free_off) {
 645                                 /*
 646                                  * Fixups XXX - big-blocks already allocated as part
 647                                  * of the freemap bootstrap.
 648                                  */
 649                                 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
 650                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
 651                                 layer2->bytes_free = 0;
 652                         } else if (phys_offset + block_offset < vol->vol_free_end) {
 653                                 layer2->zone = 0;
 654                                 layer2->append_off = 0;
 655                                 layer2->bytes_free = HAMMER_BIGBLOCK_SIZE;
 656                                 ++count;
 657                                 ++layer1_count;
 658                         } else {
 659                                 layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
 660                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
 661                                 layer2->bytes_free = 0;
 662                         }
 663                         layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 664                         buffer2->cache.modified = 1;
 665                 }
 666
 667                 layer1->blocks_free += layer1_count;
 668                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 669                 buffer1->cache.modified = 1;
 670         }
 671
 672         rel_buffer(buffer1);
 673         rel_buffer(buffer2);
 674         rel_volume(root_vol);
 675         return(count);
 676 }
 677
 678 /*
 679  * Returns the number of big-blocks available for filesystem data and undos
 680  * without formatting.
 681  */
 682 int64_t
 683 count_freemap(struct volume_info *vol)
 684 {
 685         hammer_off_t phys_offset;
 686         hammer_off_t vol_free_off;
 687         hammer_off_t aligned_vol_free_end;
 688         int64_t count = 0;
 689
 690         vol_free_off = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 691         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
 692                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
 693
 694         if (vol->vol_no == HAMMER_ROOT_VOLNO)
 695                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
 696
 697         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 698              phys_offset < aligned_vol_free_end;
 699              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 700                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
 701         }
 702
 703         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 704              phys_offset < aligned_vol_free_end;
 705              phys_offset += HAMMER_BIGBLOCK_SIZE) {
 706                 if (phys_offset < vol_free_off) {
 707                         ;
 708                 } else if (phys_offset < vol->vol_free_end) {
 709                         ++count;
 710                 }
 711         }
 712
 713         return(count);
 714 }
 715
 716 /*
 717  * Format the undomap for the root volume.
 718  */
 719 void
 720 format_undomap(struct volume_info *root_vol, int64_t *undo_buffer_size)
 721 {
 722         const int undo_zone = HAMMER_ZONE_UNDO_INDEX;
 723         hammer_off_t undo_limit;
 724         hammer_blockmap_t blockmap;
 725         struct hammer_volume_ondisk *ondisk;
 726         struct buffer_info *buffer = NULL;
 727         hammer_off_t scan;
 728         int n;
 729         int limit_index;
 730         uint32_t seqno;
 731
 732         /* Only root volume needs formatting */
 733         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
 734         ondisk = root_vol->ondisk;
 735
 736         /*
 737          * Size the undo buffer in multiples of HAMMER_BIGBLOCK_SIZE,
 738          * up to HAMMER_UNDO_LAYER2 big-blocks.  Size to approximately
 739          * 0.1% of the disk.
 740          *
 741          * The minimum UNDO fifo size is 500MB, or approximately 1% of
 742          * the recommended 50G disk.
 743          *
 744          * Changing this minimum is rather dangerous as complex filesystem
 745          * operations can cause the UNDO FIFO to fill up otherwise.
 746          */
 747         undo_limit = *undo_buffer_size;
 748         if (undo_limit == 0) {
 749                 undo_limit = (ondisk->vol_buf_end - ondisk->vol_buf_beg) / 1000;
 750                 if (undo_limit < 500*1024*1024)
 751                         undo_limit = 500*1024*1024;
 752         }
 753         undo_limit = (undo_limit + HAMMER_BIGBLOCK_MASK64) &
 754                      ~HAMMER_BIGBLOCK_MASK64;
 755         if (undo_limit < HAMMER_BIGBLOCK_SIZE)
 756                 undo_limit = HAMMER_BIGBLOCK_SIZE;
 757         if (undo_limit > HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2)
 758                 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2;
 759         *undo_buffer_size = undo_limit;
 760
 761         blockmap = &ondisk->vol0_blockmap[undo_zone];
 762         bzero(blockmap, sizeof(*blockmap));
 763         blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 764         blockmap->first_offset = HAMMER_ZONE_ENCODE(undo_zone, 0);
 765         blockmap->next_offset = blockmap->first_offset;
 766         blockmap->alloc_offset = HAMMER_ZONE_ENCODE(undo_zone, undo_limit);
 767         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
 768
 769         limit_index = undo_limit / HAMMER_BIGBLOCK_SIZE;
 770         assert(limit_index <= HAMMER_UNDO_LAYER2);
 771
 772         for (n = 0; n < limit_index; ++n) {
 773                 ondisk->vol0_undo_array[n] = alloc_bigblock(root_vol,
 774                                                         HAMMER_ZONE_UNDO_INDEX);
 775         }
 776         while (n < HAMMER_UNDO_LAYER2) {
 777                 ondisk->vol0_undo_array[n++] = HAMMER_BLOCKMAP_UNAVAIL;
 778         }
 779
 780         /*
 781          * Pre-initialize the UNDO blocks (HAMMER version 4+)
 782          */
 783         printf("initializing the undo map (%jd MB)\n",
 784                 (intmax_t)(blockmap->alloc_offset & HAMMER_OFF_LONG_MASK) /
 785                 (1024 * 1024));
 786
 787         scan = blockmap->first_offset;
 788         seqno = 0;
 789
 790         while (scan < blockmap->alloc_offset) {
 791                 hammer_fifo_head_t head;
 792                 hammer_fifo_tail_t tail;
 793                 int isnew;
 794                 int bytes = HAMMER_UNDO_ALIGN;
 795
 796                 isnew = ((scan & HAMMER_BUFMASK64) == 0);
 797                 head = get_buffer_data(scan, &buffer, isnew);
 798                 buffer->cache.modified = 1;
 799                 tail = (void *)((char *)head + bytes - sizeof(*tail));
 800
 801                 bzero(head, bytes);
 802                 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
 803                 head->hdr_type = HAMMER_HEAD_TYPE_DUMMY;
 804                 head->hdr_size = bytes;
 805                 head->hdr_seq = seqno++;
 806
 807                 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
 808                 tail->tail_type = HAMMER_HEAD_TYPE_DUMMY;
 809                 tail->tail_size = bytes;
 810
 811                 head->hdr_crc = crc32(head, HAMMER_FIFO_HEAD_CRCOFF) ^
 812                                 crc32(head + 1, bytes - sizeof(*head));
 813
 814                 scan += bytes;
 815         }
 816         rel_buffer(buffer);
 817 }
 818
 819 /*
 820  * Flush various tracking structures to disk
 821  */
 822 void
 823 flush_all_volumes(void)
 824 {
 825         struct volume_info *vol;
 826
 827         TAILQ_FOREACH(vol, &VolList, entry)
 828                 flush_volume(vol);
 829 }
 830
 831 void
 832 flush_volume(struct volume_info *volume)
 833 {
 834         struct buffer_info *buffer;
 835         int i;
 836
 837         for (i = 0; i < HAMMER_BUFLISTS; ++i) {
 838                 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
 839                         flush_buffer(buffer);
 840         }
 841         if (writehammervol(volume) == -1)
 842                 err(1, "Write volume %d (%s)", volume->vol_no, volume->name);
 843 }
 844
 845 void
 846 flush_buffer(struct buffer_info *buffer)
 847 {
 848         struct volume_info *vol;
 849
 850         vol = buffer->volume;
 851         if (writehammerbuf(buffer) == -1)
 852                 err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
 853         buffer->cache.modified = 0;
 854 }
 855
 856 /*
 857  * Core I/O operations
 858  */
 859 static int
 860 __read(struct volume_info *vol, void *data, int64_t offset, int size)
 861 {
 862         ssize_t n;
 863
 864         n = pread(vol->fd, data, size, offset);
 865         if (n != size)
 866                 return(-1);
 867         return(0);
 868 }
 869
 870 static __inline int
 871 readhammervol(struct volume_info *vol)
 872 {
 873         return(__read(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
 874 }
 875
 876 static __inline int
 877 readhammerbuf(struct buffer_info *buf)
 878 {
 879         return(__read(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
 880 }
 881
 882 static int
 883 __write(struct volume_info *vol, const void *data, int64_t offset, int size)
 884 {
 885         ssize_t n;
 886
 887         if (vol->rdonly)
 888                 return(0);
 889
 890         n = pwrite(vol->fd, data, size, offset);
 891         if (n != size)
 892                 return(-1);
 893         return(0);
 894 }
 895
 896 static __inline int
 897 writehammervol(struct volume_info *vol)
 898 {
 899         return(__write(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
 900 }
 901
 902 static __inline int
 903 writehammerbuf(struct buffer_info *buf)
 904 {
 905         return(__write(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
 906 }
 907
 908 int64_t init_boot_area_size(int64_t value, off_t avg_vol_size)
 909 {
 910         if (value == 0) {
 911                 value = HAMMER_BOOT_NOMBYTES;
 912                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
 913                         value >>= 1;
 914                 if (value < HAMMER_BOOT_MINBYTES)
 915                         value = 0;
 916         } else if (value < HAMMER_BOOT_MINBYTES) {
 917                 value = HAMMER_BOOT_MINBYTES;
 918         }
 919
 920         return(value);
 921 }
 922
 923 int64_t init_mem_area_size(int64_t value, off_t avg_vol_size)
 924 {
 925         if (value == 0) {
 926                 value = HAMMER_MEM_NOMBYTES;
 927                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
 928                         value >>= 1;
 929                 if (value < HAMMER_MEM_MINBYTES)
 930                         value = 0;
 931         } else if (value < HAMMER_MEM_MINBYTES) {
 932                 value = HAMMER_MEM_MINBYTES;
 933         }
 934
 935         return(value);
 936 }